EE ZOOM SEMINAR:
השתתפות בסמינר תיתן קרדיט שמיעה – עפ"י רישום שם + מספר ת.ז. בצ'אט
Join Zoom Meeting
https://us04web.zoom.us/j/656977141?pwd=c0ZPbTVHd1MxY3A5RW5JbFI5Ykhtdz09
Meeting ID: 656 977 141
Password: 535131
Speaker: Tomer Raviv
M.Sc. student under the supervision of Prof. Yair Be’ery
SUNDAY, April 19th, 2020 at 15:00
ZOOM
Data Driven Decoding for Error Correction Codes
Abstract
High quality data is essential in deep learning to train a robust model. While in other fields data is sparse and costly to collect, in the field of error correction codes it is free to query and label thus allowing potential data exploitation.
In this seminar, I'll explore the deep connection between data and machine-learning based decoding. After presenting the current state of the field, the adoption of ensembles for decoding applications is considered.
Ensemble models are widely used in machine learning to solve complex tasks by their decomposition into multiple simpler tasks, each one solved locally by a single member of the ensemble. However, for decoding applications, solely optimizing performance is inadequate; as one must take complexity into account as well. We suggest a low-complexity scheme, referred to as a data-driven ensemble, where only a single member participates in the decoding of each word. Every member is a weighted belief propagation decoder, formed by the parameterization and training of a classical belief propagation decoder.
Our novel scheme incorporates domain knowledge by several means. First, the distribution of feasible words is partitioned into non-overlapping regions. Thereafter, specialized experts are formed by independently training each member on a single region. A classical hard-decision decoder is employed as the gating function, mapping every word to a single expert in an injective manner. This gating introduces a priori knowledge, exploited by the ensemble at inference. Lastly, we present FER gains in the waterfall and error-floor regions for two short-length BCH codes with cycles-reduced parity-check matrices.
